Post on 22-Feb-2016
description
Bayesian SPARROW Model
Song QianIbrahim Alameddine
The University of ToledoAmerican University of Beirut
SPARROW• SPARROW: SPAtially Referenced Regressions
On Watershed attributes• SPARROW estimates the origin and fate of
contaminants in river networks• It is a semi-empirical non-linear model• It is spatial in structure and takes into
account the nested configuration of monitoring stations in a basin
• Can be used to link changes in the watershed to changes in water quality
SPARROW EQUATIONNutrient loading (L) at a downstream water
quality monitoring station i:
i
N
n iJj
Rji
Sjijnni HHeSL
)log()log(1 )(
,,,)jαZ(
# of sources
# of upstream reaches
Contribution fromDifferent sources (S)Losses/sinks
Multiplicativeerror term
SPARROW ShortcomingsSome of the shortcomings of SPARROW:• Temporal and Spatial average• Coarse spatial resolution regional specifics
often omitted• Spatial autocorrelation in model residuals• Model developed to run under
What Did We Do?• We changed the model’s architecture to make it
temporally dynamic• We developed a new regionalizing approach
– Substitute space (# of stations) with time (# of years)
• We nested the model within a larger scale regional model
• We assessed changes in loading over time for the Neuse subwatersheds
• We moved the model to an open source platform
Neuse SPARROW: Bayesian, Dynamic, & Regional
• Nested the model within the lager scale Nitrogen Southeast model (Hoos & McMahon, 2009)
• Updated the model over time (time step = 2 years)– Used 12 years of data Regionalization over time– Data and model parameters change over time
(dynamic)– Bayesian updating
(posterior of t-1 = prior at t)
How Did the Neuse BSPARROW Model Perform
Over Time?
1 2 3
4 5 6
Neu
se S
PARR
OW
: Mod
el F
it
90-91 92-93 94-95
96-97 98-99 00-01
How Do We Compare to the SE Model?
(Hoos & McMahon, 2009)
Where Are the Areas of Concern?
Have They Changed Over Time?
1990
Neuse Nitrogen Export by Basin
2001
Yield to Neuse Estuary by Basin
1990 2001
Durham
CaryMorrisville
Raleigh
Kinston
Durham
CaryMorrisville
Raleigh
Kinston
Conclusions• Regionalization of SPARROW to basin level
possible:Bayesian temporally dynamic nested modeling framework
• Loads (and model coefficients) across the basin change over time and the model is capturing these changes
• Urban runoff seems to be a concern for TN loading in the Neuse
• Nitrogen loading to the Neuse Estuary have decreased relative success in environmental management